Re: [gentoo-server] requirements for a more stable portage tree

[Dormando]

Hey,

Just read the last flurry of mails and will try to respond with my own
opinions:

I'm not a big fan of the "Gimp 1.2 -> 1.3 -> 1.4" style version tracking.
It will add a lot of overhead, and point-releases with new problems are
what we're desperately trying to avoid in the first place. Often times,
stable releases of "major" revisions (IE: BIND 8 -> BIND 9) provide much
less potential problems after initial migration. You also end up with a
system much like redhat's. Half of it is more up to date than it should
be, and the other half is frustratingly missing very helpful features.

The idea of having a stable branch for enterprise installs isn't just so
"once it's set up it'll be stable", but remember that you're not always
the only person working on your network. Most of my coworkers here operate
totally on their own. I would much prefer that they take a general build
of gentoo from whereever, and have the same install as all of the systems
I set up. Yes, I can set up a big complicated gold server with my own
mirrors and such, but if they ever need something outside of that, it'll
get ugly. Not only that, but often folks can't even budget the time to set
things up right in the first place (including myself!). This
gives people with anal-retentive PHB's a chance to put a stable
gentoo-based operating system on their servers without a huge amount of
work. Since it's perfectly possible to "freeze" your own stable tree and
maintain it given current tools, the whole frozen tree idea wouldn't be on
the table if there were outside requirements like this.

Supporting ebuilds two years back with backports? I'm not sure. We have a
*lot* of major databases here running solaris 2.6. Even with the latest
patch clusters, latest "versions" of everything for solaris 2.6... They
run with considerably more problems than the sol 8 or sol 9 machines. Even
though basic bug and security fixes are backported, there are still major
flaws in the system which were fixed by doing *major revisions*. A year is
much more acceptable without crufting the system beyond usefulness. If you
really think your sol 2.6 box is all that hot, I can pretty much assure
you, you aren't using it much.

Three year release cycles are beyond most major scopes of usefulness for
anything other than appliances and possibly firewalls. I would much rather
have focus on frozen quarterly branches with major stability
fixes/security updates. That's the big thing, in my humble opinion, which
would help break gentoo into the true enterprise of people having not
enough time and admins not always being coordinated to the fullest.

So yeah. Quarterly releases, security patches/backports up to a year back
at most, and whatever else was listed. I can't remember the rest of the
original e-mail anymore.

have fun,
-Dormando